Autonomous Execution
Auteryn agents plan and carry out multi-step tasks without you guiding each step. They use their Integrations and Knowledge and work in their sandbox, streaming results back as they go.
Execution limits: Fast runs time out at about 11 minutes. Deep work is not tied to one run. The agent checkpoints after each step, and the next run continues from the plan, notes and files the last one left, so a task can go on for days or weeks. If a run is interrupted, recovery context puts its plan and scratchpad into the next message.
Execution modes
Section titled “Execution modes”| Mode | Default for | Timeout (approx.) | Best for |
|---|---|---|---|
| Auto | Internal agents | That of the mode it picks | Let Auteryn choose between speed and depth |
| Fast | Customer-facing | ~11 min | Chat, lookups, quick automation |
| Deep | Internal (when selected) | Continues across runs | Research, multi-step coding, complex workflows |
Customer-facing agents always use Fast mode.
Auto mode
Section titled “Auto mode”When an Internal agent is set to Auto, Auteryn classifies each incoming message and runs it Fast or Deep automatically. Override to Fast or Deep in the agent’s Model section, under How hard it thinks, if you want consistent behavior.
What happens during a run
Section titled “What happens during a run”- Agent receives your message (Console, API, Loop, or Channel)
- Retrieves relevant knowledge chunks (if attached)
- Plans steps and takes actions (Integrations, sandbox, browser, web search)
- Streams its text and each action back as it happens
- On completion, async jobs may update conversation summary and memory
Recovery context
Section titled “Recovery context”If a run is interrupted (timeout, cancellation, a restart) or fails, the next message can include recovery context, which gives the agent back its previous plan and notes along with the results of its last few steps.
This is within-conversation resumption — not multi-day task persistence across unrelated sessions.